MACRO Framework Enables Task-Specific Transformer Layer Routing

MACRO implements Markov Chain Routing of Transformer Layers to learn task-specific execution paths without modifying underlying model weights. The framework supports skip, repeat, and residual operations using a context-dependent Markov policy decoded via a top-k Viterbi algorithm.